Generative design of novel metal nodes and topologies for MOFs

Develop generative artificial intelligence models for metal-organic frameworks that can de novo generate novel inorganic secondary building units (metal nodes with specific coordination environments) and novel topological nets, despite current limitations in structural data quality and coverage; specifically, design methods that enable generation beyond known building blocks and nets under datasets that often lack experimental validation and may contain structural errors.

Background

The Perspective emphasizes that available structural datasets for metal-organic frameworks present significant limitations for training generative models. Hypothetical MOF libraries provide breadth but often lack experimental validation, while experimentally rooted databases may have coverage gaps and can include structural errors such as incorrect metal oxidation states.

Within this context, the authors explicitly identify a key challenge: enabling generative models to produce genuinely novel inorganic nodes and framework topologies, rather than relying only on motifs present in existing data. This problem is central to pushing discovery beyond the current training distributions and requires approaches that are robust to data quality issues.